Models

    Noema 2B

    A 2-billion-parameter model that holds its own against far larger open models — and stays accurate at half the token budget, on your own hardware.

    Released
    July 6, 2026
    Size
    2B parameters
    Architecture
    Hybrid transformer — Gated DeltaNet + gated attention
    Description

    Noema 2B is our first release: a compact, instruction-tuned language model built on Qwen3.5-2B and post-trained to be sharper and dramatically more token-efficient. It runs entirely on your own hardware in non-thinking mode, and it is engineered for the trade every local model has to make — quality against footprint. It broadly outperforms its base and, on held-out verifiable prompts, matches or beats an open model more than four times its size.

    Token-efficient reasoningMath & step-by-step reasoningInstruction followingOn-device, non-thinking
    Parameters
    2.0B
    Base model
    Qwen3.5-2B
    Architecture
    Hybrid — Gated DeltaNet + attention
    Context length
    262K tokens
    Modes
    Non-thinking · thinking
    Formats
    SafeTensors · GGUF
    Precision
    F16 · Q8_0 · Q6_K · Q4_K_M
    License
    Apache 2.0

    Improvements

    Where Noema 2B improves on Qwen3.5-2B, its base

    +10.8 ptsMATH-500 @2048

    Math at a deployment budget

    66.2 vs 55.4 for the base at a 2,048-token budget — the setting small models actually run in.

    +7.5 ptsGSM8K · SC@8

    Grade-school math

    +3.3 single-shot and +7.5 with self-consistency; pass@8 reaches 96.7%.

    +8 ptsIFEval

    Instruction following

    Prompt-strict 73 vs 65, with gains across all four IFEval metrics — and 22% fewer tokens per response.

    +14.7 ptsHumanEval

    Code that runs

    HumanEval 53.7 vs 39.0 and HumanEval+ 50.0 vs 36.0 — markedly more correct, runnable code than the base.

    −30%Tokens / correct

    Fewer tokens per answer

    394 vs 565 tokens per correct answer on IFEval; wrong answers also fail shorter instead of running to the cap.

    2 / 120Non-termination

    Reliable generation

    Runaway, repetitive generation is effectively gone — failures terminate with clean, parseable answers.

    Benchmarks

    Higher is better · non-thinking mode

    84%

    Punches above its weight

    On 130 held-out verifiable prompts never seen in training, Noema 2B matches or beats Qwen3.5-9B — an open model over four times its size — on 84% of items.

    95.7%vs 82.4% for the base

    Budget-robust

    Share of full-budget MATH-500 accuracy retained at half the token budget. Noema stays near its ceiling; the base loses nearly a fifth of its performance when tokens are tight.

    Versus the base — and a 9B reference

    Noema 2B
    Qwen3.5-2B (base)
    Qwen3.5-9B (reference)

    GSM8K

    k=1
    NoemaBaseQ3.5-9B

    GSM8K

    SC@8
    NoemaBaseQ3.5-9B

    MATH

    100-set
    NoemaBaseQ3.5-9B

    MBPP-100

    pass@1
    NoemaBaseQ3.5-9B

    IFEval

    prompt-strict
    NoemaBaseQ3.5-9B

    All three scored on the same frozen harness (non-thinking) — identical eval for each. Qwen3.5-9B, 4.5× Noema's size, is a reference ceiling: Noema 2B closes much of the base → 9B gap.

    Coding — versus the base

    Noema 2B
    Qwen3.5-2B (base)

    HumanEval

    pass@1
    NoemaBase

    HumanEval+

    pass@1
    NoemaBase

    MBPP

    pass@1
    NoemaBase

    MBPP+

    pass@1
    NoemaBase

    Noema 2B vs Qwen3.5-2B on our frozen harness — non-thinking, pass@1. EvalPlus suite: HumanEval / MBPP plus their extended-test (+) variants. (Current general models publish LiveCodeBench, not EvalPlus, so there is no fair current-peer set here.)

    Versus the current 2–4B class

    Noema 2B
    LFM2-2.6B (Liquid)
    SmolLM3-3B

    IFEval

    NoemaLFM2SmolLM3

    GSM8K

    NoemaLFM2SmolLM3

    MATH-500

    NoemaLFM2SmolLM3

    Non-thinking mode for all. LFM2-2.6B from the LFM2 technical report (arXiv:2511.23404; no reasoning mode). SmolLM3-3B from community /no_think evals — GSM8K flexible-extract, IFEval prompt-level loose (strict not published); MATH-500 pending. Qwen3.5-4B non-thinking pending. Noema on our frozen harness (MATH-500 at a 2,048-token budget). Protocols differ across labs.